Core Minutes 11/27/2012

• ScienceTools: (Jim) refactored the code for the binned likelihood analysis to address reports from Jean Ballet regarding excessively long execution times for models with larger numbers of fixed parameters. With these changes, Jean found that performance has improved significantly. All of the recent changes that were in ST-LATEST have been moved to ST HEAD, and a release tag will be made in the next couple of weeks.

  See also Science Tools Development Notes.

• Overlay events: (Leon) The idea is to make the run-by-run generation of overlay events automatic, so we can collect a large enough sample as the amount of data increases. We currently have no idea how much is enough, but we're pretty sure that we want more than the 21M events that we have now! At the same time, we want to add a few more FT2 variables to the overlay event (instantaneous livetime, lat config/mode, data quality) so we can better monitor things, and potentially make a better selection of overlay events.

  A tiny bit of progress: I talked with Tom G. before I went on vacation. The project should proceed in 3 phases:

  1. a pipeline task to process all the runs up to 'now'. We need to decide which version of the merit files to produce. This could give us ~10X the number of events we have now.
  2. incorporation of the task into level1 (stay with pass7 or wait for pass8?)
  3. catch-up for the time between 'now' and the start of level1

  I'm working up a coherent proposal now.

• Pass 8 (Tracy) doesn't have much to report; waiting for Heather.

  (Heather) The new ROOT (v5.34.03; make that v5.34.03-gr01 for Windows) has been built for rhel5 32-bit and vc90 and there are versions of gaudi and TMine built against that ROOT version. This is all used by HEAD 1473. Systest mostly look good, but three of them are taking much too long (did not complete in extra-long queue) to run for no discrenible reason; presumably it has something to do with the new ROOT. She will try to run by hand to see what's going on. Meanwhile, Windows developers who want to get started can

  1. Grab the following externals from /nfs/farm/g/glast/u35/externals/Windows-i386-32bit-vc90:
     • ROOT-v5.34.03-gr01-dbg.zip
     • gaudi-v21r7-gl4-dbg.zip
     • TMineExt-3.3.2-gr02-dbg.zip
  2. Check out HEAD 1473 from CVS, e.g.
     cvs co -r GlastRelease-HEAD-1-1473 GlastRelease-scons
  3. See GR Windows Recipe for instructions on how to build minimal targets with GoGui or SCons directly, then bring up Visual Studio.

  (Johan) requests a rhel5 optimized build of this release. (Heather) Ok, but recently changes have been pretty much for the benefit of Windows only.

• Core talk:

  • Debug python for Windows (Heather) managed to make a debug build of 2.7.2 (clearly not a typical thing to do and then made a ROOT debug build using it which mostly succeeded, but there was this internal dll created that remained non-debug. Rather than try to fish that out. I retreated to grabbing ROOT's binary installation — which is built against python 2.6.5 — so we'll retreat to python 2.6.5 on windows. Strangely, they don't seem to build against a debug python.. but then again we don't link against the python ROOT libs in GR.. so I'm going to suggest we don't need a python debug on windows.. at the very least I don't think it's a sustainable way to go. The only known downside is that we can't make swig wrapper libraries in debug builds. This is of no importance for GR.

  • SCons RM and GlastRelease (Heather) has not been having fun trying to get RM to make GR builds on Linux. Presumably this is due to contention in the CVS server: near-simultaneous check-outs for different OSes, LATEST versus HEAD, etc. all needing to read the same files and, in some cases, writing copies to the same disk. (Joanne) suggested turning off automated builds for OSes which are less critical; it could at least decrease the frequency of problems. (Heather) We could dispense with vc71 and rhel6 for GR (these problems don't seem to occur for ST) and build them manually as needed. [Post-meeting Tom S. proposed several additional easy-to-implemnt changes to RM which should lessen contention considerably.]

  • Developer environments (Joanne) updated the GR Windows Recipe page and will add a link to it in the Workbook SCons section.

    She spent some time attempting to revise GoGui so that it would not go through the separate init step (invoking SCons with the --interactive flag so that subsequent builds don't redo computation of dependencies) any longer. This was desirable at one time, but now that SCons has improved and machines have gotten faster it's no longer a boon. Not surprisingly, the first trial run of the revised code failed (hung up). It's hard to know how much more effort would be required to get it to work so, unless there are interested customers for this feature, I'll put it aside for now.

    Left-over news from late October which never made it into minutes: as of SConsFiles tag SConsFiles-01-06-00 supersede + non-installed headers should be fully supported on Windows.

previous

minutes index

next

J. Bogart, Last Modified: